Robust Regression Discontinuity Design
Robust Bias-Corrected Regression Discontinuity Design · Also known as: Robust RDD, Bias-corrected RDD, CCT estimator, rdrobust
Robust RDD extends the classical regression discontinuity design with bias correction and robust confidence intervals, addressing the under-coverage problem of conventional RDD inference. Developed by Calonico, Cattaneo, and Titiunik (2014), it uses local polynomial estimation with a bias-corrected point estimate and a wider variance term that accounts for the added uncertainty, yielding confidence intervals with correct asymptotic coverage.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use robust RDD when you have a continuous running variable, a known cutoff that determines treatment, and you want interval estimates with correct coverage rather than merely point identification. It is especially important when the optimal bandwidth is not very small (so residual bias is non-trivial) or when reviewers or journals require bias-corrected inference. Do not use when the running variable is manipulable near the cutoff (density test should be checked first), when the design is geographically or temporally complex enough to require extensions (spatial or fuzzy RDD), or when the sample near the cutoff is very small (fewer than ~30 units per side), which makes all RDD-based inference unreliable regardless of bias correction.
Strengths & limitations
- Corrects the coverage failure of conventional RDD confidence intervals, which systematically under-cover the true parameter.
- Data-driven MSE-optimal bandwidth selection removes ad hoc choices that can be gamed to produce significant results.
- Implemented in widely available, peer-reviewed software (rdrobust in R and Stata), facilitating replication and transparency.
- Preserves the clean causal identification of standard RDD while providing honest uncertainty quantification.
- Accommodates covariates, clustered errors, and fuzzy designs within the same robust framework.
- Robust confidence intervals are wider than conventional ones by construction, which may reduce apparent precision even when the estimate is more honest.
- The method estimates a local average treatment effect only at the cutoff; external validity to units far from the threshold requires additional assumptions.
- Requires a sufficient density of observations near the cutoff; sparse data near c makes all local estimates unreliable.
- The bias correction itself introduces estimation error, so very small samples can make the correction noisy rather than helpful.
Frequently asked
Why are robust RDD confidence intervals wider than conventional ones?
Because the conventional interval assumes the bandwidth-induced bias is exactly zero, which it is not. The robust interval honestly accounts for the uncertainty in estimating and correcting that bias, making it wider but correctly calibrated — it actually contains the true parameter at the stated confidence level.
How does bias correction differ from simply choosing a smaller bandwidth?
Choosing a smaller bandwidth reduces bias but increases variance. Bias correction leaves the bandwidth unchanged for the point estimate, explicitly estimates and removes the bias with a secondary fit, and then adjusts the confidence interval width to reflect the uncertainty introduced. It achieves better coverage without forcing an excessively narrow bandwidth.
What software implements robust RDD?
The rdrobust package, available for both R and Stata, implements the full Calonico-Cattaneo-Titiunik estimator including MSE-optimal bandwidth selection, bias correction, robust standard errors, and covariate-adjustment extensions.
Should I always prefer robust RDD over classical RDD?
For published empirical work, yes. The bias-corrected robust estimator dominates the conventional one in terms of coverage accuracy at little cost. Classical (non-corrected) RDD confidence intervals are known to under-cover and are increasingly considered inadequate by referees in economics and political science journals.
Can robust RDD be applied to a fuzzy design?
Yes. The rdrobust framework extends to fuzzy RDD, where treatment receipt is not deterministic at the cutoff, by using the discontinuity in treatment probability as an instrument for treatment receipt. The same bias-correction and robust inference machinery applies.
Sources
- Calonico, S., Cattaneo, M. D., & Titiunik, R. (2014). Robust Nonparametric Confidence Intervals for Regression-Discontinuity Designs. Econometrica, 82(6), 2295-2326. DOI: 10.3982/ECTA11757 ↗
- Cattaneo, M. D., Idrobo, N., & Titiunik, R. (2019). A Practical Introduction to Regression Discontinuity Designs: Foundations. Cambridge University Press. ISBN: 978-1108710206
How to cite this page
ScholarGate. (2026, June 3). Robust Bias-Corrected Regression Discontinuity Design. ScholarGate. https://scholargate.app/en/causal-inference/robust-regression-discontinuity-design
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Difference-in-DifferencesEconometrics↔ compare
- Fuzzy Regression DiscontinuityCausal inference↔ compare
- Instrumental Variables in Health ResearchHealth Economics↔ compare
- Propensity Score MatchingResearch Statistics↔ compare